Liveness/activity snapshot for the idle scaler.

Returns how long the api has gone without a real HTTP request (health probes and this endpoint excluded) plus the number of live WebSocket connections. The in-cluster idle-scaledown CronJob reads this (in-cluster, so it reaches this VPC-only router directly) and scales staging to zero once idle_seconds crosses its threshold and the queues are drained. active_websockets is informational only — a socket left open idle must not keep staging alive — so it does not gate scale-down.